Financial Statement Fraud Detection Using Optimized Deep Neural Network

It is very difficult to detect FSF (Financial Statement Fraud) in the financial reports of organizations. Many researchers have worked on quantitative data and developed many models to detect the financial frauds in the statements but it is very difficult to detect when the numeric data is managed deliberately. The solution to this is text mining which can analyze unstructured texts. In this regard, many deep learning methods have been used to detect financial fraud. The Harris Hawks Optimization (HHO) algorithm is used in this paper to select the features, and a deer hunting optimization (DHO) is used with a deep neural network to determine whether a financial statement contains a fraudulent or non-fraudulent report. At first, text pre-processing is done then Count-Vectorizer (CV), TF-IDF (Term Frequency Inverse Document Frequency), and Word2Vec methods are employed for text feature extraction. The simulation results show that the proposed approach provides high detection accuracy (95%) with the feature selection process on testing data as compared to other standard classifiers. This method also gives better results on training and testing data using feature selection and without a feature selection process as compared to other classifiers.